Pricing
Perplexity Search API Pricing Depends on Which API You Call
Perplexity's Search API costs $5 or $1 per 1,000 successful requests; Agent bills tools plus tokens. What 1K to 1M requests cost, and the Sonar change.
Perplexity’s raw Search API costs $5 per 1,000 successful standard requests or $1 per 1,000 Fast requests, without an additional raw-search token charge. Agent API is a different budget: actual search/fetch tool invocations plus the selected model’s tokens. I would choose the output first and calculate the bill second. Pricing, Fast Search.
Short on time?
Here's my take
- Retrieved results
- Raw Search
- Standard is $5 and Fast $1 per 1,000 successful requests, without an extra raw-search token bill.
- Existing Sonar code
- Review migration
- Check sync/streaming reformulation and migrate unsupported async usage.
Prices verified . Sponsors cannot buy a verdict or a position.
What does Perplexity Search cost? #
Raw Search costs $5 per 1,000 successful standard requests or $1 for Fast, without an additional raw-search token charge. Agent adds separate search/fetch tool fees and model tokens. These USD inputs were verified 2026-11-10 and exclude taxes and account-specific agreements. Pricing, Search quickstart, Agent quickstart.
| Product / operation | Standard | Fast | Add model tokens? |
|---|---|---|---|
| Raw Search | $5 / 1,000 successful requests | $1 / 1,000 successful requests | No additional raw-search token bill |
| Agent web_search | $2.50 / 1,000 invocations | $1 / 1,000 invocations | Yes |
| Agent fetch_url | $0.50 / 1,000 invocations | No separate Fast rate listed | Yes |
Inputs: Perplexity pricing, Fast mode. A request, query, tool invocation and answer are different units.
The same page lists other Agent tools with their own fees: image_search at $2.50 per 1,000 successful calls, people_search and finance_search at $5 per 1,000, and sandbox at $0.03 per session plus search queries made inside it. Pricing.
And at 1K, 100K or 1M requests? #
At 1,000, 100,000 and one million successful requests, standard Search is $5, $500 and $5,000, while Fast is $1, $100 and $1,000. These are illustrative multiplications, not observed invoices or evidence that the modes return equivalent results. Pricing.
| Successful raw requests | Standard | Fast |
|---|---|---|
| 1,000 | $5 | $1 |
| 100,000 | $500 | $100 |
| 1,000,000 | $5,000 | $1,000 |
Assumptions: the chosen raw endpoint and mode, no extra product usage, no unverified free credit and no tax. I did not purchase credits, inspect an invoice or compare response quality.
Do extra queries and empty results count? #
A raw Search request can contain one query or up to five query strings, sharing one successful-request charge. Rate limiting counts query units separately, so a five-query request does not buy five times the documented throughput. The response supports up to 20 results with documented content and filtering controls. Search quickstart, rate limits.
Successful empty results are billable; invalid requests, 429s and upstream failures are not billed under the documented raw Search rule. I would log both success status and query count instead of treating “no useful answer” as an automatic refund. Pricing.
The raw Search limit is 50 query units per second across tiers. Agent uses separate organization-scoped rolling-window limits and tier conditions; buying enough raw requests does not establish unlimited Agent concurrency. Usage tiers.
How do I request Fast Search? #
Set search_type to fast in a raw Search request to /search. The following is an unrun reference example requiring an authorized account and a privately stored API key; it contains no actual key. Search quickstart, Fast mode.
curl https://api.perplexity.ai/search \
-H "Authorization: Bearer $PERPLEXITY_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"query":["MCP authorization specification","MCP tools output schema"],"search_type":"fast","max_results":5}'search_type: fast selects the raw Search mode; it is not the Agent fast effort level. The documentation says the field needs Python SDK 0.43.6 or later or TypeScript SDK 0.38.6 or later, and gives workarounds for older versions, so I would check the selected SDK before relying on it. Fast Search, Agent presets.
How much do Perplexity Agent tools cost? #
For 1,000 hypothetical Agent answers, each using exactly one standard search and one fetch, the tool subtotal is 1,000 × ($0.0025 + $0.0005) = $3. With one Fast search and one fetch it is $1.50. Both examples still need the selected model’s token bill, and a real agent may use more or fewer tools. Pricing, Agent web search.
Agent uses input and typed output rather than the older messages/choices assumptions. The preset parameter is now effort (the old name remains a deprecated alias), and the current levels are fast, low, medium, high, xhigh and ultra. Fast through xhigh default web_search to Fast Search. The model and search choices can change, so I would not quote a fixed all-in answer price without recording them. Quickstart, presets, Agent web search, changelog.
The official MCP also separates raw perplexity_search from Agent-backed perplexity_ask, perplexity_reason and perplexity_research. A shared integration endpoint does not make these the same billing product. MCP documentation.
What changed for Sonar? #
Sonar support ended September 27, 2026, but that does not establish a blanket outage for every older request. The migration guide describes gradual synchronous/streaming reformulation by model and says async usage is unsupported and must migrate to the appropriate background workflow. Migration guide.
I would inventory endpoint, request schema, streaming/async behavior and selected effort level before changing an existing application. A current Search request budget should not be copied from an older Sonar consumer/model table. Migration, current pricing.
Do you get free credits? And what about privacy? #
I could not confirm a recurring free API credit, including from a consumer Pro subscription, so I would budget from zero. Perplexity documents prepaid credits, optional auto-reload and administrator billing controls; an empty balance blocks API keys. I would check the actual dashboard rather than subtract an assumed benefit. Projects and billing.
The privacy documentation describes no training by default, while zero-data-retention requires eligibility, agreement, enablement and applicable provider constraints. The API terms (last updated January 23, 2026) say customer content is not used to train generative AI models, but third-party model providers’ terms also apply and I did not complete a legal review, so this summary does not make it ready for confidential production data. Privacy/security, API terms.
What would I compare it with? #
I would compare Exa for source retrieval, Tavily for its credit-metered search/extraction path, and Brave for its Search output. OpenAI/Claude built-in search and Gemini grounding buy different model-integrated workflows, so tool rates alone do not settle total cost. Exa, Tavily, Brave, OpenAI, Claude, Gemini.
For Google organic/SERP rows, SerpentAPI at apiserpent.com has Web rates of $0.60/$0.30/$0.10/$0.05/$0.03 per 1,000 units at Default/Growth/Scale/Enterprise/Enterprise Plus, with the paid tiers reached by deposits of $100/$500/$1,000/$1,250; your best deposit in a calendar month sets the price for that month’s deposits. Deep bills requested pages, and credit you already hold keeps the price you bought it at. That low consumption price does not turn it into an Agent answer or page-extraction service. Pricing, docs, FAQ, terms.
The search API pillar, my side-by-side web search API prices, Exa pricing, Tavily pricing and built-in search decision guide cover those separate decisions.
FAQ #
How much does the Perplexity Search API cost? #
Standard costs $5 per 1,000 successful requests and Fast $1, without an additional raw-search token charge. Perplexity pricing.
Is Search the same as Sonar or Agent? #
No. Search returns retrieved results; Agent generates model output with separately billed tools and tokens. Sonar has its own migration requirements. Perplexity Agent quickstart.
Do five queries in one request cost five times as much? #
Up to five query strings can share one successful raw Search request charge, but rate limiting counts the individual query units. Perplexity Search quickstart.
Do I pay for empty results or errors? #
Successful empty results are billed. The documented raw Search rule excludes invalid requests, 429s and upstream failures. Perplexity pricing.
Is Fast Search the same as the Agent fast preset? #
No. search_type fast selects raw Search mode. Agent fast is a separate effort level (formerly called a preset) with model and tool choices. Perplexity Fast Search.
Does Perplexity Pro come with free API credit? #
This research did not verify a recurring free API allowance from a consumer subscription. Check the actual API project balance and billing settings. Perplexity projects and billing.
Sources
23 sources, accessed .
Official pricing, docs, changelogs and status pages 23
- Perplexity pricing
- Perplexity Search quickstart
- Perplexity Fast Search
- Perplexity Agent quickstart
- Perplexity Sonar migration
- Perplexity presets
- Perplexity Agent web search
- Perplexity projects and billing
- Perplexity rate limits
- Perplexity privacy and security
- Perplexity API terms
- Perplexity changelog
- Perplexity API MCP
- SerpentAPI pricing
- SerpentAPI FAQ
- SerpentAPI API documentation
- SerpentAPI terms
- Tavily credit documentation
- Exa pricing
- Brave API pricing
- OpenAI pricing
- Claude web search
- Gemini pricing
